FI-KAN: Fractal Interpolation Kolmogorov-Arnold Networks

TL;DR

FI-KAN introduces learnable fractal bases into Kolmogorov-Arnold Networks, significantly improving approximation of non-smooth and fractal functions by adapting basis complexity to target regularity.

Contribution

The paper proposes FI-KAN, integrating fractal interpolation functions into KAN, enabling adaptive multi-scale basis functions that match target regularity for improved function approximation.

Findings

Hybrid FI-KAN outperforms KAN across regularity levels by 1.3x to 33x.

FI-KAN reduces MSE by up to 6.3x on fractal targets.

Hybrid FI-KAN achieves up to 79x improvement on rough PDE solutions.

Abstract

Kolmogorov-Arnold Networks (KAN) employ B-spline bases on a fixed grid, providing no intrinsic multi-scale decomposition for non-smooth function approximation. We introduce Fractal Interpolation KAN (FI-KAN), which incorporates learnable fractal interpolation function (FIF) bases from iterated function system (IFS) theory into KAN. Two variants are presented: Pure FI-KAN (Barnsley, 1986) replaces B-splines entirely with FIF bases; Hybrid FI-KAN (Navascues, 2005) retains the B-spline path and adds a learnable fractal correction. The IFS contraction parameters give each edge a differentiable fractal dimension that adapts to target regularity during training. On a Holder regularity benchmark ($\alpha \in [0.2, 2.0]$), Hybrid FI-KAN outperforms KAN at every regularity level (1.3x to 33x). On fractal targets, FI-KAN achieves up to 6.3x MSE reduction over KAN, maintaining 4.7x advantage at 5 dB SNR. On non-smooth PDE solutions (scikit-fem), Hybrid FI-KAN achieves up to 79x improvement on rough-coefficient diffusion and 3.5x on L-shaped domain corner singularities. Pure FI-KAN's complementary behavior, dominating on rough targets while underperforming on smooth ones, provides controlled evidence that basis geometry must match target regularity. A fractal dimension regularizer provides interpretable complexity control whose learned values recover the true fractal dimension of each target. These results establish regularity-matched basis design as a principled strategy for neural function approximation.